Optimizing Chromatographic Separation with Redosing: Effects on Separation Efficiency of a Model System in Centrifugal Partition Chromatography

Abstract

This study investigates and optimizes chromatographic separation in a Centrifugal Partition Chromatograph. Therefore, a model system is separated in a single-disc rotor. The occurring loss of the stationary phase lowers the separation efficiency over time. We introduced a new mode of operation, called the redosing of the stationary phase, to counteract this hydrodynamic phenomenon. Experiments with redosing at an optimized operating point demonstrate almost constant separation performance over 12 h, reducing solvent consumption by 45% and increasing chromatographic resolution by 37%. The improvement in retention by 69% contributes to this enhancement. Accordingly, reference experiments without redosing were conducted as a benchmark, highlighting the automated mode’s benefits, as mentioned.

1. Introduction

Biotechnology plays a decisive role in the challenging task of decarbonization on an industrial scale. It enables alternative synthesis routes based on renewable raw materials that are otherwise only accessible via petrochemistry and fossil materials. Biotechnological processes are already widely used today, especially in manufacturing active pharmaceutical ingredients, such as vaccines, drugs, or antibiotics [1,2,3]. They offer further advantages like reduced toxicity and the benefit of reduced greenhouse gas emissions [4,5].

Although the biotechnological production of pharmaceutical products has many advantages, the challenges must be addressed. One of the biggest tasks in the production of pharmaceuticals is the purification of target components, which underlies high-quality requirements and has to be very efficient since the concentrations of products are typically low [6,7,8]. As a result, downstream processing in pharmaceutical processes accounts for the largest share of costs, so developing new processes and optimizations is of great interest [9].

A suitable method for this task is chromatography, a thermal process that works under gentle conditions and is often non-degrading. A typical chromatographic setup consists of a mobile and an immobilized stationary phase. The mobile phase can be a liquid or a gas, while the stationary phase is often a solid component. The stationary phase is filled into a column and circumvented by the mobile phase. Although a solid phase is frequently chosen as the stationary phase, this approach comes with a few disadvantages, like high costs for the columns and a loss of product caused by irreversible adsorption at the stationary phase [5,10].

One approach to overcome these disadvantages is liquid–liquid chromatography (LLC), which uses two immiscible liquids instead of one fluid and one solid component. The separation is based on the different affinity of the target components to each liquid. Among others, the main advantage is that no irreversible adsorption occurs, which allows for full recovery and a high selectivity due to the many different biphasic systems that can be utilized [11,12,13]. As an example, research addresses the tetrahydrocannabinol remediation of hemp extracts utilizing LLC, optimizing biphasic solvent systems and operating parameters [14,15].

After separation into time-shifted fractions, the resulting chromatogram can be analyzed with the help of the resolution R, which measures separation efficiency (i.e., baseline separation of peaks). In LLC, the resolution can be calculated according to Equation (1) [16].

$$\mathrm{R}=Sf·\frac{\sqrt{{\mathrm{N}}_{\mathrm{theo}}}}{4}·\frac{\left({\mathrm{K}}_{\mathrm{D},2}-{\mathrm{K}}_{\mathrm{D},1}\right)}{\left\{1-Sf·\left[1-\frac{{\mathrm{K}}_{\mathrm{D},2}+{\mathrm{K}}_{\mathrm{D},1}}{2}\right]\right\}}$$

(1)

N_theo is the number of theoretical stages, and Sf is the retention value [17]. The partition coefficient K_D,i of a target molecule to be separated can be calculated according to Equation (2), with c_i,stat and c_i,mob being the equilibrium concentration in the stationary and mobile phase.

$${\mathrm{K}}_{\mathrm{D},\mathrm{i}}=\frac{{\mathrm{c}}_{\mathrm{i},\mathrm{stat}}}{{\mathrm{c}}_{\mathrm{i},\mathrm{mob}}}$$

(2)

Parameters affecting the separation process in general are the sample properties (i.e., partition coefficients), the physical properties of the biphasic system used (densities, interfacial tension, viscosities), as well as the operational parameters of the apparatus (volumetric flow rate, rotational speed, mode of operation, rotor geometry) [18,19,20,21]. For example, high volumetric flow rates promote highly dispersed hydrodynamics, increasing the rate of the loss of the stationary phase over time (as explained in the following). Therefore, an optimum has to be found: a given separation task needs a suitable biphasic system and corresponding operational parameters of the plant to ensure stable hydrodynamics and sufficient separation efficiency [22].

Hence, one main challenge in operating LLC is immobilizing the liquid stationary phase. One possible setup to solve this task is Centrifugal Partition Chromatography (CPC), invented in 1982 by Muroyama et al. In this setup, chambers connected by ducts are arranged circularly around a rotor, which spins around the central axis. The liquid stationary phase is immobilized with the help of the resulting centrifugal field. Computational Fluid Dynamics simulations show that the mobile phase is pumped through the cascade of chambers and channels and disperses at each chamber inlet before coalescing near the outlet [23,24].

Since LLC is based on the affinity of the components to each phase, it is essential to have a high mass transfer, which goes hand in hand with sufficient dispersion and, therefore, a maximized interfacial area. However, this is limited to the extent that an increase in dispersion leads to less coalescence in the chambers since the interfacial area increases, and the coalescence of the dispersed drops takes more time [25]. This causes the entrainment of drops in the stationary phase and, subsequently, the loss of the stationary phase, again decreasing the separation efficiency. This effect is also referred to as bleeding [16,17].

Our recent research focuses on counteracting bleeding by redosing the stationary phase [17,26,27]. Therefore, the amount of stationary phase lost during a specific timespan is redosed into the apparatus repeatedly. We were able to stabilize hydrodynamics and, therefore, consistently maintain a constant phase ratio inside the rotor. These novel findings could contribute to the future industrial relevance of LLC. The work reported in this paper highlights the effect of redosing the stationary phase on separating components. Thus, a suitable model system, containing D(+)-carvone and salicylic acid, was selected, and corresponding hardware (i.e., detector) was installed. As a biphasic system, the Arizona N system was chosen (1:1:1:1 volumetric amounts of water, ethyl acetate, n-heptane, and methanol) [28]. In previous publications, we conducted intensive research on the behavior and properties of different solvent systems of the Arizona series [22,29,30]. Therein, the Arizona N system has been studied most, including, among others, flow regimes and the optimization of the chamber design [17,22,23,30]. Another advantage of the Arizona N system is its straightforward handling, and the ratio between the upper and lower phase (0.7) is comparably similar [28]. This is a benefit since further lower phases are needed during the operation in descending mode (ρ_{mobile phase} > ρ_{stationary phase}).

In addition, the software was upgraded to implement automated sample addition. In the second step, additional tests were carried out to tune the algorithm, which goes hand in hand with optimizing the redosing procedure. The operability was subsequently examined and compared with operation without redosing.

To conclude, the scientific questions to be addressed here are as follows:

• Separation efficiency with redosing of the stationary phase.
  The influence of the redosing process on the separation must be addressed. With stabilized retention values over time, the chromatographic resolution should be constant too. Therefore, separation runs without redosing are necessary as a benchmark. Furthermore, potential optimization of the separation should be highlighted.
• Interdependence of redosing and sample injection.
  It is essential to analyze whether the redosing of the stationary phase and the injection of samples during operation interfere.

2. Materials and Methods

2.1. Phase System and Model System

Arizona N, no older than 36 h, was used for all experiments. Depending on the expected consumption of solvent, different amounts of the Arizona N system were prepared by mixing purified water (by MILLI-Q^® system, Millipak^® Express 40, Merck, Darmstadt, Germany), methanol (99%), ethyl acetate (99.9%), and n-heptane (99.8%) (all supplied by VWR International, Radnor, PA, USA) in equal volumetric amounts. The chemicals were directly mixed and equilibrated for 30 min and tempered in a water bath at 21 °C during the experiments. Depending on the experiments conducted, the lower aqueous phase was stained to enhance the contrast of the two phases, which was needed for the camera-based evaluation of stationary phase retention. Therefore, methylene blue was used (20 mg·L⁻¹_{lower phase}).

We defined several requirements to find suitable sample components to separate: They should not be safety-critical substances to ensure easy handling. In addition, substances with industrial relevance are preferred. Furthermore, the substances should be detectable with different detectors to maintain a broad selection possibility for this sensor. Ideally, the substances should also be available at short notice and low cost.

The G.U.E.S.S. mix offers a selection of natural substances of various polarities [16]. We reviewed which substances fulfilled the previously mentioned requirements. Then, the approximate resolution was calculated according to Equation (1) to identify suitable components for separation in a single-disc rotor with 66 chambers. It is assumed that each chamber equals a theoretical separation stage N and that the volume of the stationary phase is equal to that of the mobile phase [5]. A resolution of R ≥ 1.5 is desired to ensure baseline separation of the resulting peaks. Literature data showed a partition coefficient for D(+)-carvone in the Arizona N system of K_D = 5 and K_D = 1 for salicylic acid, giving a theoretical resolution of R = 2 [22].

This value is above the optimum of R ≈ 1.5 [5], fulfilling the requirements. However, the number of theoretical separation stages N may be less than the number of chambers, and bleeding would increase the ratio of mobile to stationary phase, resulting in a loss of resolution. Nevertheless, the model system consisting of D(+)-carvone (98%) and salicylic acid (99.9%) (all supplied by VWR International, Radnor, PA, USA) was utilized for all separation experiments. A defined volume of the lower phase was transferred to a bottle and mixed with the sample substances to prepare the sample components.

Before installation of the detector in the chromatographic plant, preceding experiments were conducted. The spectra (200 to 450 nm) for both molecules were measured with the help of a diode-array UV detector. The data recorded were analyzed with the software Clarity Version 8.7.19 from DataApex [31]. The pure aqueous phase of Arizona N was used as a reference, so only the absorption of D(+)-carvone or salicylic acid was measured.

2.2. Centrifugal Partition Chromatograph

The plant setup is given schematically in Figure 1 and was used for all experiments. A detailed description of the devices is additionally provided in

Table 1. Overview of the devices in the experimental setup.

Device	Manufacturer	Model
CCD camera	Jai Pulnix, Yokohama, Japan	AccuPIXEL TM 1327 GE
chromatograph	Kromaton, Annonay, France	FCPC-A 30506
controller unit	Gardasoft, Cambridge, UK	RTCC 420
detector	Knauer, Berlin, Germany	Smartline DA-UV-detector
phase pumps	Knauer, Berlin, Germany	P 4.1S with 50 mL pump head
sample pump	Ismatec, Stadtprozelten, Germany	MS-4/12 100 Reglo Digital
fractionation valve	Knauer, Berlin, Germany	HighSpeedValve
six-port-two-position valves	Knauer, Berlin, Germany	K-6/12/16

. The main element of the experimental setup is the FCPC rotor (66 chambers, Sf_max = 0.782, V_rotor = 10.4 mL), which is connected to the rest of the plant via two rotary joints [17]. A high-speed camera is placed over the rotor to record the chambers in the viewing windows. The camera is connected to a computer, which processes the recorded material using a MATLAB algorithm and can be used to control all peripherals using LabVIEW [17,27,32,33]. Below the rotor, a light fork barrier is installed, triggering the flash during the recordings.

Both phases are pumped through a degasser to a first six-way two-position valve with two single pumps. The valve position defines which phase is pumped back to the reservoir and which is pumped to the second valve containing a sample loop. The sample pump, connected to a separate sample reservoir, can automatically fill the sample loop. Depending on the position of the valve, the sample is transported to the rotor, or the pure mobile phase is pumped to the rotor. After the injection of the sample, the position is changed back, and the loop is refilled with the sample. Behind the detector, an eight-port one-way valve is installed, determining whether the fluid is pumped to the fractionating tanks or back to the reservoir. If the Arizona N system is contaminated with the sample, it is pumped into the fractionation system to avoid an accumulation of the sample in the process. If not, the solvent system is recycled back to the reservoir.

All experiments are carried out in descending mode. Therefore, the lower heavy phase is used as the mobile phase, and the upper light phase is stationary. When starting an experiment, the rotor is filled with the stationary phase first, and the inlet flow is subsequently switched to the mobile phase flow. The main variables to be evaluated are the chromatographic resolution and the retention over time.

2.3. Influence of Redosing on Resolution

Concerning the complete automation of the plant, the extent to which a redosing of the stationary phase influences the separation must be investigated. Thus, 200 mg·L⁻¹ salicylic acid and 200 mg·L⁻¹ D(+)-carvone were dissolved in the heavy phase of Arizona N. The sample injection volume was set to V_inj = 0.9 mL.

To investigate the influence of redosing, two worst-case scenarios are considered, i.e., redosing of the stationary phase directly before or after a sample injection. The redosing volumes were chosen to be of a similar order of magnitude to the volumes of redosing volumes in previous investigations [26]. The rotational speed was set to 750 rpm, and the volume flow of the mobile phase was set to 15 mL·min⁻¹.

2.4. Optimizing the Resolution

An optimized operating point for separating D(+)-carvone and salicylic acid is found in the next step. The influence of the retention on the resolution was analyzed for different volumetric flow rates of the mobile phase (5, 10, 15, 20 mL·min⁻¹) at 750 rpm and V_inj = 0.9 mL. The Sf*_setpoint was increased from Sf = 0 to the equilibrium Sf* in steps of 0.1, resulting in the operation points shown in

Table 2. Operating points for investigating the separation performance at different volume flows and setpoints for the stationary phase retention. The highest setpoint for each flow rate is determined by the equilibrium Sf* [26].

Volumetric Flow Rate of Mobile Phase [mL·min−1]	Sf*setpoint
5	{0.26, 0.35, 0.45, 0.55, 0.65, 0.75, 0.85, 0.9}
10	{0.26, 0.35, 0.45, 0.55, 0.65, 0.75}
15	{0.26, 0.35, 0.45, 0.55, 0.65}
20	{0.26, 0.35, 0.45, 0.55}

[26]. To ensure detection of the samples in the linear range of the detector, concentrations were kept at 200 mg·L⁻¹ per component.

2.5. Separation Experiments with Redosing

The experimental routine for the separation experiments is shown in Figure 2. Each cycle starts with a waiting time t_loop, which has to elapse before the first Sf* measurement is performed. The files are automatically stored and subsequently processed by MATLAB, as discussed previously [17,27]. The outcome is then transferred to the controller, which needs a setpoint for the phase retention Sf*_setpoint as a second input. The controller’s output is the stationary phase volume to be redosed V_stat, which is then added to the rotor. Subsequently, a specific time t_ctrl has to elapse before a control measurement is performed. After the control measurement, a sample can be injected, which is then separated during the waiting time for the next cycle t_loop. This routine is repeated for n cycles.

The operating point for the separation experiments was a volumetric flow rate of the mobile phase of 10 mL·min⁻¹, a rotational speed of 750 rpm, and concentrations of 200 mg·L⁻¹ for each component to be separated. As discussed in our previous publication, t_ctrl should equal six dimensionless residence times of the rotor [26]. Furthermore, after sample injection, t_loop is set to 9 min to ensure maximized productivity. A new cycle starts as soon as possible after the sample has left the rotor. The Sf*_setpoint is set to 0.72.

2.6. Reference Experiments without Redosing

Finally, a reference experiment was carried out in which the sample was injected at the same interval as in the long-term test, whereby redosing of the stationary phase was omitted. The test was carried out until the retention fell below the critical retention of Sf = 0.2, defined as the lower retention limit for sufficient separation in general [16]. Everything was prepared in the same way as for the separation experiments with redosing, and the setpoint was set to Sf*_setpoint = 0.1 to avoid any redosing. Accordingly, all measurements were carried out identically; the only difference was that no stationary phase was redosed at all.

3. Results and Discussion

3.1. Model System Analysis

The spectra of D(+)-carvone and salicylic acid were measured in a diode-array detector to determine suitable wavelengths for recording the chromatograms.

As can be seen in Figure 3, D(+)-carvone shows absorption between 200 nm and 280 nm, with a maximum at 239 nm. A comparison with literature data confirms the recorded spectrum [34]. The spectrum of salicylic acid shows absorption ranging from 200 nm to 345 nm. One maximum can be found at 239 nm and a second maximum at 300 nm. Between the two maxima, a local minimum at 260 nm occurs. Again, literature data confirm the recorded spectrum [35]. Since D(+)-carvone and salicylic acid show their global absorption maximum at 239 nm, this wavelength was chosen to record the chromatograms. A further wavelength to analyze was 300 nm, as salicylic acid has a second absorption maximum at this wavelength. D(+)-carvone does not absorb at this wavelength, indicating that this signal is selective for salicylic acid, provided there is no impurity. In addition, another measurement is taken at 450 nm to detect disturbances, for example, due to air bubbles, since none of the substances show absorption at this wavelength.

3.2. Influence of Redosing on Resolution

We investigated to what extent the separation in the apparatus is influenced when the stationary phase is redosed directly before or after the sample injection. The two extreme cases resulting from redosing immediately before or after sample injection are investigated and compared with separation after the disturbance caused by the redosing has subsided. The latter will be referred to as the normal separation. In this case, a range for the redosing volume is given since this experiment was conducted with an automated redosing routine (therefore, variable redosing volumes of the stationary phase). The results are shown in

Table 3. Influence of redosing on the resolution R of the separation of salicylic acid and D(+)-carvone. Measured: Redosing one second before or after injection, compared with separation after settling time of six dimensionless residence times (normal separation). Volume flow of 15 mL·min⁻¹ and rotational speed of 750 rpm. Measured at 21 ± 2 °C, in descending mode and with an Arizona N system. Injection volume V_inj = 0.9 mL, concentration of each sample component c_{salicylic acid} = c_D(+)-carvone = 200 mg·L⁻¹. The results of the one-way analysis of variance are 1.196 (T-value), 0.340 (p-value) and 2.484 (critical F-value).

	Redosing → Injection			Injection → Redosing			Normal Separation
Vstat [mL]	0.125	0.25	0.375	0.125	0.25	0.375	0–0.32
R [-]	0.615 ± 0.032	0.638 ± 0.066	0.668 ± 0.048	0.620 ± 0.041	0.624 ± 0.059	0.677 ± 0.040	0.677 ± 0.037

. The mean resolution of the tripled experiments is insignificantly different.

The results indicate that redosing the stationary phase directly before or after the injection of a sample has no significant effect on the resolution and, thus, the separation efficiency for the given system. Therefore, it is theoretically possible to decouple the control of the retention and the separation in the rotor without scheduling these two processes in future experiments. Nevertheless, since each retention measurement requires the pumps to be stopped, again influencing the detection, the timing of injection and redosing was considered for the following experiments. Otherwise, spikes occur in the chromatogram once the pumps are stopped, hindering the resolution calculation. Figure 4 shows an example chromatogram with spikes of the two retention measurements (highlighted in orange). After these two measurements, the sample is injected and separated. The third peak with absorption maxima at 239 nm and 300 nm can be assigned to salicylic acid (marked blue), while the fourth peak with a maximum only at 239 nm corresponds to D(+)-carvone (marked grey).

3.3. Optimizing the Resolution

The next goal is to optimize the operation point of the plant regarding the resolution. Therefore, the resolution was measured for different phase retentions and volumetric flow rates of the mobile phase. It is known from previous experiments that the rotational speed does not significantly influence the separation efficiency [30]. Thus, 750 rpm was the rotational speed in all cases [22,30]. The results are shown in Figure 5. The resolution is not specified for small Sf* values below 0.35, as evaluation of the resolution was impossible because of overlapping peaks. It can be seen that, irrespective of the volume flow, the resolution is of a similar order of magnitude for the given retention. It is noticeable that the standard deviation for resolution at larger retention values is significantly higher for 5 mL·min⁻¹. A possible reason is an increased signal-to-noise ratio in these measurements, leading to significant deviations in the evaluation.

The optimization aims at maximizing the resolution. Therefore, the three operating points at 5 mL·min⁻¹ with the highest chromatographic resolution and those with the highest resolution at 10 mL·min⁻¹ are considered as operating points because a statistical evaluation shows that they do not differ significantly. In addition to the resolution, other performance parameters should be examined as well. For this purpose, the separation time is defined as the time from the beginning of the first peak to the end of the second peak. The lower the separation time, the higher the productivity at constant resolution. The corresponding measurements are shown in

Table 4. Measured separation times of the separation for different volume flows at 750 rpm. The separation time is calculated at each volume flow’s maximum possible retention and resolution. Measured at 21 ± 2 °C, in descending mode and with an Arizona N system.

Volumetric Flow Rate of Mobile Phase [mL·min−1]	Separation Time [min]
5	$13.89\pm 0.49$
10	$7.54\pm 0.30$
15	$4.38\pm 0.23$
20	$3.42\pm 0.19$

, indicating that the separation time decreases with increasing volume flow, which is to be expected. However, it is noticeable that the separation time does not scale proportionally to the volume flow. This may be because the flow regime (i.e., the degree of dispersion) changes over the volume flow, influencing the separation time [17].

The maximal retention and, therefore, the resolution decrease with increasing volumetric flow rates, as mentioned and shown in Figure 5. At 20 mL·min⁻¹, the highest retention reached was 0.51 ± 0.0008, and the corresponding resolution was 0.66 ± 0.007, whereas the best-case retention at 10 mL·min⁻¹ was 0.75 ± 0.001 and the resolution 0.94 ± 0.04.

On the other hand, the signal-to-noise ratio must be considered. As mentioned, the noise significantly increases for a volume flow of 5 mL·min⁻¹. Concerning the selection of an optimized operating point, a volume flow of 10 mL·min⁻¹ is selected, as this offers a shorter separation time combined with less noise and an optimized separation performance.

In conclusion, optimizing the operating point results in the following parameters: a volumetric flow rate of 10 mL·min⁻¹, 750 rpm, and an Sf*_setpoint of 0.72. The setpoint is set slightly below the maximum possible Sf*_equilibrium because the setpoint should always be reached within the redosing interval, which makes slight overdosing (induced by the shifting factor) necessary [26]. If the setpoint was set identical to the equilibrium point, this overdosing would not be possible or would be associated with increased solvent consumption without an increase in retention.

3.4. Separation Experiments with Redosing

The results of the experiments at the optimized operation point are shown in Figure 6.

Over the entire duration of the experiment, the average phase retention is 0.71 ± 0.01, indicating an insignificant offset from the setpoint (Sf*_setpoint = 0.72). Therefore, redosing the stationary phase can compensate for bleeding in this case. Further, 0.11 ± 0.006 rotor volumes per hour were added to maintain the phase retention. Compared to the data described in our previous publication, the consumption is approximately twice as high due to the increased bleeding caused by sample injection [26]. The average resolution is 0.82 ± 0.02. The results show that due to the new mode of operation, it is possible to operate the chromatograph stably and constantly over a prolonged period. At the same time, the redosing of the stationary phase is capable of compensating for the hydrodynamic influences of sample injection automatically.

The number of stages in the apparatus can be calculated based on the measured resolution and the phase retention. The partition coefficients are K_{D,salicylic acid} = 1 and K_{D,D(+)-carvone} = 5 [22]. According to Equation (1), the total number of stages is N_theo = 9.93 ± 0.35. This number is a measure for the column (or rotor) efficiency. With 66 chambers in the rotor, the number of stages per chamber equals N_{theo per chamber} = 0.15 ± 0.0053. As discussed in the literature, the difference in the number of theoretical separation stages is rather low compared to liquid chromatography applications. Nevertheless, this downside is compensated for by the much higher stationary phase volume when comparing liquid chromatography and liquid–liquid chromatography [16].

An excerpt of a chromatogram from the experiments is given in Figure 7. As described before, the spikes can be attributed to the stationary phase retention measurements. After the measurements with maxima at both wavelengths, the first peak corresponds to salicylic acid, while the second peak with a single maximum at 239 nm belongs to D(+)-carvone.

In conclusion, the optimal operation point regarding the chromatographic resolution is at a volumetric flow rate of 10 mL·min⁻¹, at 750 rpm, and at Sf*_setpoint = 0.72. Samples of 200 mg·L⁻¹ salicylic acid and D(+)-carvone can be separated with a mean resolution of 0.82.

3.5. Reference Experiments without Redosing

As explained, a reference test was carried out without redosing. The results are shown in Figure 8. It can be seen that without redosing of the stationary phase, the retention decreases constantly, resulting in a decreasing chromatographic resolution.

The average resolution is 0.60 ± 0.07 (compared to an average resolution of 0.82 ± 0.02 in the experiments with redosing). After completion of a reference experiment without redosing, the rotor has to be refilled entirely with the stationary phase before a new separation run can be started (dashed vertical line in Figure 8). The duration of one experimental run (corresponding to the number of sample injections in one run) is limited by a critical lower chromatographic resolution, which has to be defined for the specific separation task. For the experimental runs shown in Figure 8, the lower retention limit was defined as Sf = 0.2/Sf* = 0.26. After refilling, a subsequent separation run without redosing starts (Figure 8: transparent yellow and red data).

The consumption of the stationary phase can be calculated using a volume balance, resulting in 0.2 rotor volumes per hour. Therefore, approximately twice the amount of solvent (stationary phase) is needed for the reference runs compared to the experiments with redosing, where the average stationary phase consumption was determined to be 0.11 rotor volumes per hour. Nevertheless, all resolution values measured indicate peak overlapping (as shown in Figure 4: baseline separation not achieved). In addition to being acceptable for research purposes (cuts in the resolution are acceptable, as this is the only way to utilize a transparent rotor with corresponding online retention measurement), compromises in purity are not acceptable for industrial applications. Therefore, the number of theoretical stages should be increased to optimize the separation efficiency [36,37].

All in all, the redosing of the stationary phase as a newly developed mode of operation offers some benefits compared to different approaches, addressing the loss of the stationary phase over time. Sequential Centrifugal Chromatography or Continuous Centrifugal Extraction both circumvent the drawbacks in separation efficiency caused by bleeding but require high technical effort for implementation as well as complex operation [22,30,38,39,40].

4. Conclusions

This work investigated and optimized the chromatographic separation in a Centrifugal Partition Chromatograph by redosing the stationary phase. The separation of salicylic acid and D(+)-carvone in Arizona N was studied as a model system. Investigations were carried out in a single-disc rotor with 66 chambers. This rotor has viewing windows, allowing for an online retention measurement. To maintain constant retention values, the stationary phase was redosed utilizing a closed-loop controller. With subsequent sample injection, constant chromatographic resolution over time was achieved. The formation of other compounds during the separation was investigated (see Supplementary Materials).

In addition, an optimized operating point with maximized resolution was found. The operability with almost constant separation performance was demonstrated over about 12 h of operating time. The separation performance was kept constant, and the solvent consumption was nearly halved (45% decrease) simultaneously. The chromatographic resolution could be increased by 37% (from 0.60 to 0.82). The improvement in retention grew by 69% (from 0.42 to 0.71), achieved through redosing, which was responsible for this improvement. With this, the number of theoretical stages per chamber utilizing closed-loop redosing is 0.15 for the given model separation task. With 66 chambers in total, the number of theoretical stages equals 9.93.

The necessary scale-up already mentioned can be achieved by scaling up the rotor: several rotor discs with embedded chambers and channels can be stacked to increase the overall number of chambers in the system. The individual discs are sealed with interlayers made from polytetrafluoroethylene foil. Regrettably, this also precludes the possibility of online measurement of the rotor’s retention values using a camera, which is why the reduced chromatographic resolution was accepted in this study. It is necessary, therefore, to develop a customized online sensor for bleeding measurements to enable the scale-up of the rotor and to measure retention values in multi-disc rotors. With such a system, constant resolution values would enable the optimization of the operating point to minimize the safety buffer, further optimizing the efficiency.